DOCSLIB.ORG

Sport Analytics

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Sport Analytics SPORT ANALYTICS Dr. Jirka Poropudas, Director of Analytics, SportIQ [email protected] Outline 1. Overview of sport analytics • Brief introduction through examples 2. Team performance evaluation • Ranking and rating teams • Estimation of winning probabilities 3. Assignment: ”Optimal betting portfolio for Liiga playoffs” • Poisson regression for team ratings • Estimation of winning probabilities • Simulation of the playoff bracket • Optimal betting portfolio 11.3.2019 1. Overview of sport analytics 11.3.2019 What is sport analytics? B. Alamar and V. Mehrotra (Analytics Magazine, Sep./Oct. 2011): “The management of structured historical data, the application of predictive analytic models that utilize that data, and the use of information systems to inform decision makers and enable them to help their organizations in gaining a competitive advantage on the field of play.” 11.3.2019 Applications of sport analytics • Coaches • Tactics, training, scouting, and planning • General managers and front offices • Player evaluation and team building • Television, other broadcasters, and news media • Entertainment, better content, storytelling, and visualizations • Bookmakers and bettors • Betting odds and point spreads 11.3.2019 Data sources • Official summary statistics • Aggregated totals from game events • Official play-by-play statistics • Record of game events as they take place • Manual tracking and video analytics • More detailed team-specific events • Labor intensive approach • Data consistency? • Automated tracking systems • Expensive • Consistency based on given event definitions 11.3.2019 Data sources • Official summary statistics • Aggregated totals from game events • Official play-by-play statistics • Record of game events as they take place • Manual tracking and video analytics • More detailed team-specific events • Labor intensive approach • Data consistency? • Automated tracking systems • Expensive • Consistency based on given event definitions 11.3.2019 Data sources • Official summary statistics • Aggregated totals from game events • Official play-by-play statistics • Record of game events as they take place • Manual tracking and video analytics • More detailed team-specific events • Labor intensive approach • Data consistency? • Automated tracking systems • Expensive • Consistency based on given event definitions 11.3.2019 Data sources • Official summary statistics • Aggregated totals from game • https://www.youtube.com/edit?vide events o_id=7IdxFcy3PFA • Official play-by-play statistics • Record of game events as they take place • Manual tracking and video analytics • More detailed team-specific events • Labor intensive approach • Data consistency? • Automated tracking systems • Expensive • Consistency based on given event definitions 11.3.2019 Methodology • Basic statistics and more advanced techniques • Signal vs. noise • Mathematical modeling • Rules and scoring system specific factors • Machine learning • Neural networks, deep learning, Bayesian networks etc. • Optimization • Simulation 11.3.2019 EPL (football) – Expected goals http://www.bbc.com/sport/football/40699431 11.3.2019 NHL (ice hockey) M.B. McCurdy, @ineffectivemath, https://twitter.com/i/web/status/899721405083906048. 11.3.2019 NBA (basketball) – Houston Rockets K. Goldsberry, Grantland.com, http://grantland.com/the-triangle/future-of-basketball-james-harden-daryl-morey-houston-rockets/. 11.3.2019 MLB (baseball) – Launch angle and velocity D. Sheinin and A. Emamdjomeh, Washington Post, https://www.washingtonpost.com/graphics/sports/mlB-launch-angles-story/. 11.3.2019 NFL (American football) – 4th Down Bot B. Burke and K. Quealy, 4th Down bot, New York Times. http://www.nytimes.com/newsgraphics/2013/11/28/fourth-downs/post.html 11.3.2019 NFL (American football) – 4th Down Bot B. Burke and K. Quealy, 4th Down bot, New York Times. http://www.nytimes.com/newsgraphics/2013/11/28/fourth-downs/post.html 11.3.2019 2. Team performance evaluation and prediction of future outcomes 11.3.2019 Motivation for team performance evaluation and prediction • Unbiased evaluation of performance • Signal vs. noise • Strength of schedule • Strategy and planning • Team building and “tanking” • Storytelling and entertainment • Betting analytics • Betting lines • Predictive analytics 11.3.2019 Team performance evaluation by ranking and rating • The game results depend on (at least) three factors • Home advantage • Strength of the teams • Random variation (stochastic component) • The game results are observed and the teams are ranked or rated according to their perceived level of performance. • The objective of ranking and rating of teams is compare the underlying strengths of the teams. • Ranking: ordinal scale, i.e., the separation between successive teams is not evaluated. • Rating: interval scale, i.e., the differences between teams are measurable and have an meaningful interpretation. • Team ratings can be used for predicting the winners of future games 11.3.2019 Prediction and winning probability • Prediction of future results • When estimates for team strengths have been calculated, they can be used for estimating winning probabilities in future games. • Modeling approach depends on the rules and the scoring system • How are the points/goals scored? • Assumptions about the underlying scoring processes • N.B., There are always a number of alternative modeling choices 11.3.2019 Football • Low scoring game • Limited number of scoring chances • EPL: 2.77 goals/game in 2016-17 • Poisson distribution • Scoring intensity • “Small chance of a goal at every time instant” • Rough approximation https://dashee87.github.io/football/python/predicting-football-results-with-statistical-modelling/ 11.3.2019 Basketball • High number of scoring chances • NBA teams average ≈100 possessions per game • Consecutive offensive possessions are more or less independent • Central limit theorem • Distribution of points can be approximated using a normal distribution J. Poropudas, Kalman filter algorithm for rating and prediction in basketball, 2011. 11.3.2019 How certain is the outcome of the game? • Law of large numbers • Probability of an “upset” • In football, a match between a very good and a very bad team can still result in a tie or even an upset. • In basketball, the better team usually wins. 11.3.2019 Bradley-Terry model • Flexible model for almost(?) any game with two teams/players • Bernoulli trial: first team either wins or doesn’t. • Outcome of each game is 0 or 1. • Home advantage and scoring margin are not considered. • Parameters • Team ratings !" representing team strengths ()* • Winning probability when team # meets team $: log = !" − !/ +,()* • Parameter estimation using maximum likelihood • No closed form solution • Numerical methods 3 3 ℓ 1 = 2 2 4"/ log !" − 4"/ log !" + !/ " / E. Zermelo, Die Berechnung der Turnier-Ergebnisse als ein Maximumproblem der Wahrscheinlichkeitsrechnung, Mathematische Zeitschrift. 29 (1): 436-460, 1929. R.A. Bradley and M.E. Terry, Rank Analysis of Incomplete Block Designs: I. The Method of Paired Comparisons, Biometrika, 39 (3/4): 324-345, 1952. 11.3.2019 Maher’s model for football • “Scoring margin contains information.” • Poisson scoring for home team and visiting team: !" ~ $%&''%((*+ , -.) !0 ~ $%&''%((1. , 2+) • Four parameters per team • Offense at home and away: *+ and 1+ • Defense away and at home: -+ and 2+ • Number of parameters can decreased with equality constraints. • Parameter estimation using maximum likelihood • No closed form solution 9 9 • Numerical methods ℓ 4, 6 = 8 8 :+.*+-. − :+. log *+-. • Not a “perfect fit” to actual data + . • Independence assumption! M.J. Maher, Modelling association football scores, Statistica Neerlandica. 36 (3): 109-118, 1982. 11.3.2019 Dixon-Coles model for football • Refinement of the Maher’s model • Modification to outcomes 0-0, 1-0, 0-1, and 1-1 • Dependence between teams’ scoring • Better fit to actual results • Parameter estimation using maximum likelihood M.J. Dixon and S.G. Coles, Modelling association football scores and inefficiencies in the football betting market, Applied Statistics, 46 (2): 265-280, 1997. 11.3.2019 3. Course assignment: Optimal betting portfolio for Liiga Playoffs 11.3.2019 Finnish ice hockey league: Liiga • Top Finnish Ice Hockey League • 15 teams • 60 games for each team (30 home games) • 10 teams qualify for the playoffs • See, http://liiga.fi/ottelut/2018-2019/runkosarja/. • Regular season ends 14.3.2019 • Preliminary playoffs end 19.3.2019 • N.B., you can use all the information available up to that date in your project work. • Deadline for this project • Presentation due 1.4.2019 • Report due 13.4.2019 11.3.2019 Liiga standings (as of 10.3.2019) http://liiga.fi/tyokalut/laskuri/ 11.3.2019 Liiga playoff format • Six best teams at the conclusion of regular season proceed directly to quarter-finals • Teams placing between 7th and 10th (inclusive) will play preliminary play-offs (“wild card round”) best-of-three • The two winners of the preliminary playoffs take the last two slots to quarter-finals • All series after this are best-of-seven • In all playoff series, the team with the higher playoff seed holds the home advantage. • In the semifinals, the matchups are determined based on the regular season and the best team plays against the worst team (“re-seeding”). • N.B., you can skip the preliminary playoffs, if you like. 11.3.2019 Liiga playoffs (last season) 11.3.2019 Poisson regression • Poisson regression • Generalized linear model form of regression analysis
Load more

Recommended publications
  • Introduction Predictive Vs. Earned Ranking Methods
    An overview of some methods for ranking sports teams Soren P. Sorensen University of Tennessee Knoxville, TN 38996-1200 [email protected] Introduction The purpose of this report is to argue for an open system for ranking sports teams, to review the history of ranking systems, and to document a particular open method for ranking sports teams against each other. In order to do this extensive use of mathematics is used, which might make the text more difficult to read, but ensures the method is well documented and reproducible by others, who might want to use it or derive another ranking method from it. The report is, on the other hand, also more detailed than a ”typical” scientific paper and discusses details, which in a scientific paper intended for publication would be omitted. We will in this report focus on NCAA 1-A football, but the methods described here are very general and can be applied to most other sports with only minor modifications. Predictive vs. Earned Ranking Methods In general most ranking systems fall in one of the following two categories: predictive or earned rankings. The goal of an earned ranking is to rank the teams according to their past performance in the season in order to provide a method for selecting either a champ or a set of teams that should participate in a playoff (or bowl games). The goal of a predictive ranking method, on the other hand, is to provide the best possible prediction of the outcome of a future game between two teams. In an earned system objective and well publicized criteria should be used to rank the teams, like who won or the score difference or a combination of both.
    [Show full text]
  • Quantifying the Influence of Deviations in Past NFL Standings on the Present
    Can Losing Mean Winning in the NFL? Quantifying the Influence of Deviations in Past NFL Standings on the Present The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters Citation MacPhee, William. 2020. Can Losing Mean Winning in the NFL? Quantifying the Influence of Deviations in Past NFL Standings on the Present. Bachelor's thesis, Harvard College. Citable link https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37364661 Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Can Losing Mean Winning in the NFL? Quantifying the Influence of Deviations in Past NFL Standings on the Present A thesis presented by William MacPhee to Applied Mathematics in partial fulfillment of the honors requirements for the degree of Bachelor of Arts Harvard College Cambridge, Massachusetts November 15, 2019 Abstract Although plenty of research has studied competitiveness and re-distribution in professional sports leagues from a correlational perspective, the literature fails to provide evidence arguing causal mecha- nisms. This thesis aims to isolate these causal mechanisms within the National Football League (NFL) for four treatments in past seasons: win total, playoff level reached, playoff seed attained, and endowment obtained for the upcoming player selection draft. Causal inference is made possible due to employment of instrumental variables relating to random components of wins (both in the regular season and in the postseason) and the differential impact of tiebreaking metrics on teams in certain ties and teams not in such ties.
    [Show full text]
  • When NBA Teams Don't Want To
    GAMES TO LOSE When NBA teams don’t want to win Team X Stefano Bertani Federico Fabbri Jorge Machado Scott Shapiro MBA 211 Game Theory, Spring 2010 Games to Lose – MBA 211 Game Theory Games to lose – When NBA teams don’t want to win 1. Introduction ................................................................................................................................................. 3 1.1 Situation ................................................................................................................................................ 3 1.2 NBA Structure ........................................................................................................................................ 3 1.3 NBA Playoff Seeding ............................................................................................................................... 4 1.4 NBA Playoff Tournament ........................................................................................................................ 4 1.5 Home Court Advantage .......................................................................................................................... 5 1.6 Structure of the paper ............................................................................................................................ 5 2. Situation analysis ......................................................................................................................................... 6 2.1 Scenario analysis ...................................................................................................................................
    [Show full text]
  • Bayesian Analysis of Home Advantage in North American Professional Sports Before and During COVID‑19 Nico Higgs & Ian Stavness*
    www.nature.com/scientificreports OPEN Bayesian analysis of home advantage in North American professional sports before and during COVID‑19 Nico Higgs & Ian Stavness* Home advantage in professional sports is a widely accepted phenomenon despite the lack of any controlled experiments at the professional level. The return to play of professional sports during the COVID‑19 pandemic presents a unique opportunity to analyze the hypothesized efect of home advantage in neutral settings. While recent work has examined the efect of COVID‑19 restrictions on home advantage in European football, comparatively few studies have examined the efect of restrictions in the North American professional sports leagues. In this work, we infer the efect of and changes in home advantage prior to and during COVID‑19 in the professional North American leagues for hockey, basketball, baseball, and American football. We propose a Bayesian multi‑level regression model that infers the efect of home advantage while accounting for relative team strengths. We also demonstrate that the Negative Binomial distribution is the most appropriate likelihood to use in modelling North American sports leagues as they are prone to overdispersion in their points scored. Our model gives strong evidence that home advantage was negatively impacted in the NHL and NBA during their strongly restricted COVID‑19 playofs, while the MLB and NFL showed little to no change during their weakly restricted COVID‑19 seasons. In professional sports, home teams tend to win more on average than visiting teams1–3. Tis phenomenon has been widely studied across several felds including psychology4,5, economics6,7, and statistics8,9 among others10.
    [Show full text]
  • How Much of the NBA Home Court Advantage Is Explained by Rest?
    How Much of the NBA Home Court Advantage Is Explained by Rest? Oliver Entine Dylan Small Department of Statistics, The Wharton School, University of Pennsylvania Home Court Advantage in Different Pro Sports Home Team Winning % Basketball (NBA) 0.608 Football (NFL) 0.581 Hockey (NHL) 0.550 Baseball (Major Leagues) 0.535 Data from NBA 2001-2002 through 2005-2006 seasons; NFL 2001 through 2005 seasons; NHL 1998-1999 through 2002- 2003 seasons; baseball 1991-2002 seasons. Summary: The home advantage in basketball is the biggest of the four major American pro sports. Possible Sources of Home Court Advantage in Basketball • Psychological support of the crowd. • Comfort of being at home, rather than traveling. • Referees give home teams the benefit of the doubt? • Teams are familiar with particulars/eccentricities of their home court. • Different distributions of rest between home and away teams (we will focus on this). Previous Literature • Martin Manley and Dean Oliver have studied how the home court advantage differs between the regular season and the playoffs: They found no evidence of a big difference between the home court advantage in the playoffs vs. regular season. Oliver estimated that the home court advantage is about 1% less in the playoffs. • We focus on the regular season. Distribution of Rest for Home vs. Away Teams Days of Rest Home Team Away Team 0 0.14 0.33 1 0.59 0.47 2 0.18 0.13 3 0.06 0.05 4+ 0.04 0.03 Source: 1999-2000 season. Summary: Away teams are much more likely to play back to back games, and are less likely to have two or more days of rest.
    [Show full text]
  • Open Evan Bittner Thesis.Pdf
    THE PENNSYLVANIA STATE UNIVERSITY SCHREYER HONORS COLLEGE DEPARTMENT OF STATISTICS PREDICTING MAJOR LEAGUE BASEBALL PLAYOFF PROBABILITIES USING LOGISTIC REGRESSION EVAN J. BITTNER FALL 2015 A thesis submitted in partial fulfillment of the requirements for a baccalaureate degree in Statistics with honors in Statistics Reviewed and approved* by the following: Andrew Wiesner Lecturer of Statistics Thesis Supervisor Murali Haran Associate Professor of Statistics Honors Adviser * Signatures are on file in the Schreyer Honors College. i ABSTRACT Major League Baseball teams are constantly assessing whether or not they think their teams will make the playoffs. Many sources publish playoff probabilities or odds throughout the season using advanced statistical methods. These methods are somewhat secretive and typically advanced and difficult to understand. The goal of this work is to determine a way to calculate playoff probabilities midseason that can easily be understood and applied. The goal is to develop a method and compare its predictive accuracy to the current methods published by statistical baseball sources such as Baseball Prospectus and Fangraphs. ii TABLE OF CONTENTS List of Figures .............................................................................................................. iii List of Tables ............................................................................................................... iv Acknowledgements .....................................................................................................
    [Show full text]
  • Riding a Probabilistic Support Vector Machine to the Stanley Cup
    J. Quant. Anal. Sports 2015; 11(4): 205–218 Simon Demers* Riding a probabilistic support vector machine to the Stanley Cup DOI 10.1515/jqas-2014-0093 elimination rounds. Although predicting post-season playoff outcomes and identifying factors that increase Abstract: The predictive performance of various team the likelihood of playoff success are central objectives of metrics is compared in the context of 105 best-of-seven sports analytics, few research results have been published national hockey league (NHL) playoff series that took place on team performance during NHL playoff series. This has between 2008 and 2014 inclusively. This analysis provides left an important knowledge gap, especially in the post- renewed support for traditional box score statistics such lockout, new-rules era of the NHL. This knowledge gap is as goal differential, especially in the form of Pythagorean especially deplorable because playoff success is pivotal expectations. A parsimonious relevance vector machine for fans, team members and team owners alike, often both (RVM) learning approach is compared with the more com- emotionally and economically (Vrooman 2012). A better mon support vector machine (SVM) algorithm. Despite the understanding of playoff success has the potential to potential of the RVM approach, the SVM algorithm proved deliver new insights for researchers studying sports eco- to be superior in the context of hockey playoffs. The proba- nomics, competitive balance, home ice advantage, home- bilistic SVM results are used to derive playoff performance away playoff series sequencing, clutch performances and expectations for NHL teams and identify playoff under- player talent. achievers and over-achievers. The results suggest that the In the NHL, the Stanley Cup is granted to the winner Arizona Coyotes and the Carolina Hurricanes can both of the championship after four playoff rounds, each con- be considered Round 2 over-achievers while the Nash- sisting of a best-of-seven series.
    [Show full text]
  • Why the 2020 LA Dodgers Are the Greatest Team of All Time
    Floersch UWL Journal of Undergraduate Research XXIV (2021) Why the 2020 Dodgers Are the Greatest Team of All Time, at least statistically Sean Floersch Faculty Mentor: Chad Vidden, Mathematics and Statistics ABSTRACT This paper explores the use of sports analytics in an attempt to quantify the strength of Major League Baseball teams from 1920-2020 and then find which team was the greatest team of all time. Through the use of basic baseball statistics, ratios were personally created that demonstrate the strength of a baseball team on offense and defense. To account for slight year to year differences in the ratios, standard deviations to the yearly mean are used and the personally created Flo Strength metric is used with the standard deviations. This allows the ratios to be compared across all seasons, with varying rules, number of teams, and number of games. These ratios and Flo Strength metric go through statistical testing to confirm that the ratios are indeed standardized across seasons. This is especially important when comparing the strange Covid-19 2020 baseball season. After quantifying all the team strengths, it is argued that the 2020 Los Angeles Dodgers are the greatest team of all time. INTRODUCTION The 2020 Covid MLB Season In a year like no other, with a pandemic locking down society, racial inequalities becoming the forefront of the news, and division across the country mounting, sports were turned to for a sense of normalcy. Slowly, sports returned, but in a way never seen before. The NBA returned, isolated in a bubble. Strict protocol was put in place in the NHL and MLS in order to ensure player and personnel safety.
    [Show full text]
  • Team Payroll Versus Performance in Professional Sports: Is Increased Spending Associated with Greater Success?
    Team Payroll Versus Performance in Professional Sports: Is Increased Spending Associated with Greater Success? Grant Shorin Professor Peter S. Arcidiacono, Faculty Advisor Professor Kent P. Kimbrough, Seminar Advisor Duke University Durham, North Carolina 2017 Grant graduated with High Distinction in Economics and a minor in Statistical Science in May 2017. Following graduation, he will be working in San Francisco as an Analyst at Altman Vilandrie & Company, a strategy consulting group that focuses on the telecom, media, and technology sectors. He can be contacted at [email protected]. Acknowledgements I would like to thank my thesis advisor, Peter Arcidiacono, for his valuable guidance. I would also like to acknowledge my honors seminar instructor, Kent Kimbrough, for his continued support and feedback. Lastly, I would like to recognize my honors seminar classmates for their helpful comments throughout the year. 2 Abstract Professional sports are a billion-dollar industry, with player salaries accounting for the largest expenditure. Comparing results between the four major North American leagues (MLB, NBA, NHL, and NFL) and examining data from 1995 through 2015, this paper seeks to answer the following question: do teams that have higher payrolls achieve greater success, as measured by their regular season, postseason, and financial performance? Multiple data visualizations highlight unique relationships across the three dimensions and between each sport, while subsequent empirical analysis supports these findings. After standardizing payroll values and using a fixed effects model to control for team-specific factors, this paper finds that higher payroll spending is associated with an increase in regular season winning percentage in all sports (but is less meaningful in the NFL), a substantial rise in the likelihood of winning the championship in the NBA and NHL, and a lower operating income in all sports.
    [Show full text]
  • Home Advantage and Tied Games in Soccer
    HOME ADVANTAGE AND TIED GAMES IN SOCCER P.C. van der Kruit 0. Introduction. Home advantage and the occurence of tied matches are an important feature of soccer. In the national competition in the Netherlands about half the games end in a victory for the home team, one quarter end up tied, while only one quarter results in a win for the visiting team. In addition, the number of goals scored per game (one average about 3) is rather low. The home advantage is supposedly evened out between the teams by playing a full competition, where each two teams play two matches with both as home team in turn. I have wondered about the matter of the considerable home advantage in soccer ever since I first became interested in it.1 This was the result of the book “Speel nooit een uitwedstrijd – Topprestaties in sport en management”2 by Pieter Winsemius (1987). In this book he discusses various aspects of business management and illustrates these with facts and anecdotes from sports. In the first chapter he notes that in the German Bundesliga the top teams get their high positions in the standings on the basis of regularly winning away games. Teams on lower positions have often lost only a few more home matches, but it is the away games where they have gained much fewer points. The message is clear: never play an away game! In management terms this lesson means, according to Winsemius, for example that you arrange difficult meetings to take place in your own office. Clearly, home advantage plays an important role in soccer competitions.
    [Show full text]
  • NBA Team Home Advantage: Identifying Key Factors Using an Artificial Neural Network
    RESEARCH ARTICLE NBA team home advantage: Identifying key factors using an artificial neural network ☯ ☯ Austin R. HarrisID *, Paul J. Roebber Atmospheric Science Program, Department of Mathematical Science, University of Wisconsin±Milwaukee, Milwaukee, Wisconsin, United States of America ☯ These authors contributed equally to this work. * [email protected] Abstract a1111111111 a1111111111 What determines a team's home advantage, and why does it change with time? Is it some- a1111111111 thing about the rowdiness of the hometown crowd? Is it something about the location of the a1111111111 a1111111111 team? Or is it something about the team itself, the quality of the team or the styles it may or may not play? To answer these questions, season performance statistics were downloaded for all NBA teams across 32 seasons (83±84 to 17±18). Data were also obtained for other potential influences identified in the literature including: stadium attendance, altitude, and team market size. Using an artificial neural network, a team's home advantage was diag- OPEN ACCESS nosed using team performance statistics only. Attendance, altitude, and market size were Citation: Harris AR, Roebber PJ (2019) NBA team unsuccessful at improving this diagnosis. The style of play is a key factor in the home advan- home advantage: Identifying key factors using an artificial neural network. PLoS ONE 14(7): tage. Teams that make more two point and free-throw shots see larger advantages at e0220630. https://doi.org/10.1371/journal. home. Given the rise in three-point shooting in recent years, this finding partially explains pone.0220630 the gradual decline in home advantage observed across the league over time.
    [Show full text]
  • Using Bayesian Statistics to Rank Sports Teams (Or, My Replacement for the BCS)
    Ranking Systems The Bradley-Terry Model The Bayesian Approach Using Bayesian statistics to rank sports teams (or, my replacement for the BCS) John T. Whelan [email protected] Center for Computational Relativity & Gravitation & School of Mathematical Sciences Rochester Institute of Technology π-RIT Presentation 2010 October 8 1/38 John T. Whelan [email protected] Using Bayesian statistics to rank sports teams Ranking Systems The Bradley-Terry Model The Bayesian Approach Outline 1 Ranking Systems 2 The Bradley-Terry Model 3 The Bayesian Approach 2/38 John T. Whelan [email protected] Using Bayesian statistics to rank sports teams Ranking Systems The Bradley-Terry Model The Bayesian Approach Outline 1 Ranking Systems 2 The Bradley-Terry Model 3 The Bayesian Approach 2/38 John T. Whelan [email protected] Using Bayesian statistics to rank sports teams Ranking Systems The Bradley-Terry Model The Bayesian Approach The Problem: Who Are The Champions? The games have been played; crown the champion (or seed the playoffs) If the schedule was balanced, it’s easy: pick the team with the best record If schedule strengths differ, record doesn’t tell all e.g., college sports (seeding NCAA tourneys) 3/38 John T. Whelan [email protected] Using Bayesian statistics to rank sports teams Ranking Systems The Bradley-Terry Model The Bayesian Approach Evaluating an Unbalanced Schedule Most NCAA sports (basketball, hockey, lacrosse, . ) have a selection committee That committee uses or follows selection criteria (Ratings Percentage Index, strength of schedule, common opponents, quality wins, . ) Football (Bowl Subdivision) has no NCAA tournament; Bowl Championship Series “seeded” by BCS rankings All involve some subjective judgement (committee or polls) 4/38 John T.
    [Show full text]